Single cell clutering after regressing out cell cycle effect using seurat

Hi,

I am using cancer cell line scRNA-seq data to fing rarely expressed cells in homogenous cell culture. I am following seurat vignette for clustering. I also fund the seurat tutorial to regress out cell cycle genes effect. However, I have a basic question. My current work is as follow:

1) Find variable genes

2) Scale data

3) Performed cell cycle scoring analysis

4) Re-scale data with regressing out cell cycle genes effect

5) Run PCA using variable feature generated in step one.

But I am thinking, shouldn't I find out variable genes again if I have made some cell cycle gene based correction, and then run PCA and clustering. Some of the variable genes are cell cycle genes such as TOP2A.

Any comment will be appreciated!

seurat scrna-seq clustering

• 8,937 views • link

updated 2.8 years ago by Haci

• written 6.9 years ago by piyushjo

edited: I think the cell cycle based correction is just to filter some genes from your original HVGs. You should still have enough HVGs after filtering for clustering. If not, you could modify the threshold of step1 to get more initial HVGs.

• 0 views • link

written 6.9 years ago by shoujun.gu

Ok Thank you for your explanation.

• 0 views • link

written 6.9 years ago by piyushjo

Sorry, I made some mistake in my previous statement. Actually, it calculates a score based on the difference of mean expression of the given list and the mean expression of reference genes (randomly selected genes matching the distribution of the expression of the given list). Then regress out the scores for downstream analysis. Therefore, if you want to remove (hopefully) the cell cycle effect, you should do HVG analysis on the corrected data.

• 0 views • link

written 6.9 years ago by shoujun.gu

1 answer

If you check Seurat vignette on cycle cycle scoring (the one compiled on 2019-06-24), you will see that there are basically two steps. First, cell cycle scores (corresponding to S and G2M states) are calculated. These scores are then used to regress-out associated signal from the whole of the data. So unlike @shoujun.gu suggested, the HVGs stay the same but their signal (scaled values) change after the regression. Therefore no need to recalculate HVGs after cell cycle regression.

• 0 views • link

written 6.9 years ago by Haci

Hi Haci, I am using Seuratv3.1. I tried to regress out the cell cycle genes but nothing changed

marrow <- ScaleData(Combined, vars.to.regress = c("S.Score", "G2M.Score"), features = rownames(Combined)) marrow <- RunPCA(marrow, features = VariableFeatures(marrow), nfeatures.print = 10) marrow <- RunPCA(marrow, features = c(s.genes, g2m.genes)) DimPlot(marrow) For you it worked? Any comment will be appreciated!

• 0 views • link

written 6.9 years ago by kyriakds •

Did you PCA plot cells by there S, G2M and G1 score and saw different cluster for each stage like Fig 3 in tutorial? If yes, then it suggest that cell cycle genes have significant input to variation. If not then they don't, in that case regressing wouldn't make any difference. If you did see different clusters belonging to different stages of cell cycle, than something is wrong in your score or regression formula.

• 0 views • link

written 6.9 years ago by piyushjo